It's gradual on all variable aperture zooms.
Basically (though not exactly), the focal length changes while the pupil size remains the same. And given that the f-number is f(ocal length)/pupil size, any change in focal length, no matter how minute, will change the f-number.
And that provides the clue to how to use such a lens with non-TTL metering: first calculate the pupil size by doing the sum focal length/f-number at that focal length (always given for both extremes, so doable), and jot that down somewhere handy.
Then after changing focal length do the sum backwards again: f-number = focal length set/number calculated in first step.
What you also could do is do the sums and calculate at what focal length setting you would need to add 1/3 or 1/5 stop, and mark those points on the focal length scale of the lens.
Mind you: while close enough, it's not necessarily correct, nor exact. It depends on what happens inside that lens, and that varies with design.
But it works.
And another mind you: this changes with focus setting too. The closer you get, the more light you lose.
And that can be a significant amount not just when doing macro. Especially not when using wildly asymmeric lenses, which zooms are.